Convolution Neural Networks based lungs disease detection and Severity classification

Classification is a crucial component of Computer Aided Diagnosis (CADx) systems. This phase comprises the extraction of features. Deep features have emerged as a new topic of study in numerous disciplines, including medical imaging. However, these works contain flaws, such as excessive classification, and do not reflect the real world. This paper provides an overview of deep learning for detecting lung illness in medical photos. In the past five years, just one review article has been published on deep learning for lung illness diagnosis. We investigate utilizing deep learning to detect and categorize Convolution neural network (CNN) numerous lung illnesses from chest X-ray images. We developed a pipeline for segmenting chest X-ray (CXR) images before classification and compared the performance of our framework to that of existing techniques. To recover lung characteristics, the Binary Spotted Hyena optimizer (BSHO) was used in this study. We demonstrated that simple models and classifiers, such as shallow CNN, can compete with complex systems. Furthermore, we validated our method using publicly available lung datasets from Shenzhen and Montgomery and compared its efficacy to that of existing methods. Despite having fewer trainable parameters, our technique outperformed the top performing models trained on the Montgomery dataset in terms of accuracy. In addition, although being computationally cheaper, our CNN-BSHO model performed nearly as well as the top solution on the Shenzhen dataset. This research employed four classifiers, including Support Vector Machine (SVM), Nave Bayes, Random Forest, and Visual Geometry Group (VGG). Using CNN-BSHO, an accuracy of 98.324% was reached.